With the increasing development of modern industries, ensuring the safety and reliability of production processes becomes a more and more urgent task. Alarm root cause analysis plays a very significant role in preventing faults of complex industrial processes. Causal network modeling is an important part of alarm root cause analysis. For building a good causal network model, transfer entropy is usually adopted as an effective method. However, there are some problems in determining prediction horizons of transfer entropy. In order to solve these problems and further enhance the performance of the original transfer entropy method, a modified transfer entropy method taking the prediction horizon from one variable to another and to itself simultaneously into consideration is proposed. Moreover, based on data-driven and process knowledge based modeling methods, an approach integrating transfer entropy with superficial process knowledge is designed to correct the false calculation of transfer entropy and then optimize the causal network model to improve the capacity of causality detection. In order to verify the effectiveness of the proposed approach, two case studies using a stochastic process and a complex chemical process named Tennessee Eastman are carried out. Simulation results show that the discovering ability of causality detection can be much improved.